Claude Code 2.1.325 asks before sandboxed commands and unattended skills run

Claude Code 2.1.325 makes sandboxed Bash commands follow the normal permission flow and asks before scheduled tasks run skills with hooks. Teams with unattended runs should check.

what changed

Claude Code 2.1.325 was listed on 13 November 2026. These are the changes that matter most for teams:

  • Sandboxed Bash commands now follow the normal permission flow by default. Setting sandbox.autoAllowBashIfSandboxed: true keeps the old behaviour of running them without prompts. Managed settings that turn the sandbox on need that setting in the highest-ranking managed source.
  • Scheduled tasks and /loop ask before running a skill that declares allowed-tools or hooks. Adding Skill(name) to permissions.allow lets it run unattended.
  • “Always allow” in Desktop, IDE and SDK hosts now offers a rule only when saving it would allow that command again. A scheduled task that relied on such a rule can now wait for you.
  • Read-only git commands ask first in a repository that is not the project’s or one you trusted.
  • The sandbox refuses writes to .claude settings, skills, agents, commands and hooks in added directories.
  • Programs the sandbox starts for its own use now run without Claude Code’s sign-in tokens and API keys in their environment.

why it matters

Teams that run Claude Code on a schedule or in automation may find jobs that used to finish now stopping at a prompt. Nobody is there to answer, so the job waits. This hits sandboxed Bash, skills with hooks and saved “always allow” rules.

The other changes narrow what a sandboxed session can reach. That is good for safety. It also means a setup that wrote to .claude in an added directory may now fail.

what we would do

  • List every scheduled task, /loop and unattended run we have. Run each once after upgrading and watch for a stuck prompt.
  • Decide on purpose whether to set sandbox.autoAllowBashIfSandboxed: true. Set it only where the sandbox is the control we trust.
  • For each unattended skill we trust, add an explicit Skill(name) allow rule, and review what its hooks do first.
  • Check any workflow that writes to .claude in an added directory, and move that write to the launch folder.

sources

Drafted with Claude from the linked sources, reviewed by us.